Research map / Vision-language and multimodal / Multimodal LLMs
Dynamic MDETR: A Dynamic Multimodal Transformer Decoder for Visual Grounding
Fengyuan Shi, Ruopeng Gao, Weilin Huang, Limin Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| decouple grounding into encoder plus text guided decoder (all points) (ablation) | TransVG · RefCOCOg val-u | top-1 accuracy: 67.66 -> 69.12 | Improved |
| add 2D adaptive sampling (36 points) to the text guided decoder (ablation) | Dynamic MDETR · RefCOCOg val-u | accuracy and GFLOPS: 69.12 -> 69.20; 4.10 -> 2.23 GFLOPS | Improved |
| add text guided decoding on top of 2D adaptive sampling (ablation) | Dynamic MDETR · RefCOCOg val-u | top-1 accuracy: 67.89 -> 69.20 | Improved |
| use learnable initial sampling instead of non-learnable grid or uniform (ablation) | Dynamic MDETR · RefCOCOg val-u | top-1 accuracy: 67.77 -> 69.20 | Improved |
| use dynamic 2D adaptive sampling instead of static uniform sampling (ablation) | Dynamic MDETR · RefCOCOg val-u | top-1 accuracy: 66.81 -> 69.20 | Improved |
| add language guidance to the sampling query (ablation) | Dynamic MDETR · RefCOCOg val-u | top-1 accuracy: 68.32 -> 69.20 | Improved |
| progressively reduce sampled points (64-32-16) instead of constant 36 (ablation) | Dynamic MDETR · RefCOCOg val-u | top-1 accuracy: 69.20 -> 69.34 | Improved |
| use only 4 sampled points (1% of points) (ablation) | Dynamic MDETR · RefCOCOg val-u | top-1 accuracy: 0.22% drop | No clear effect |
| use text guided decoding instead of average pooling of sampled features (ablation) | Dynamic MDETR · RefCOCOg val-u | top-1 accuracy: 67.89 -> 69.20 | Improved |
| use text guided decoding instead of self attention decoding (ablation) | Dynamic MDETR · RefCOCOg val-u | top-1 accuracy: 68.30 -> 69.20 | Improved |
| use language queries instead of MDETR object queries (ablation) | Dynamic MDETR · RefCOCOg val-u | top-1 accuracy: 66.81 -> 69.20 | Improved |
| use encoder-decoder with 2D sampling instead of encoder-only 1D attention-based sampling | TransVG · RefCOCOg val-u | top-1 accuracy: 67.67 -> 69.20 | Improved |
Later papers that compared with this method
2 later accepted papers report Dynamic-MDETR as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- AerialVG: A Challenging Benchmark for Aerial Visual Grounding by Exploring Positional Relations (ICCV 2025 · better on 2 of 2 shared results)
- SimVG: A Simple Framework for Visual Grounding with Decoupled Multi-modal Fusion (NeurIPS 2024 · better on 5 of 5 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- TransVG++ — TransVG: End-to-End Visual Grounding With Transformers (ICCV 2021) · better on 1 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SimVG: A Simple Framework for Visual Grounding with Decoupled Multi-modal Fusion (NeurIPS 2024 · similarity 0.89)
- Referencing Where to Focus: Improving Visual Grounding with Referential Query (NeurIPS 2024 · similarity 0.87)
- GeM-VG: Towards Generalized Multi-image Visual Grounding with Multimodal Large Language Models (AAAI 2026 · similarity 0.86)
- VGent: Visual Grounding via Modular Design for Disentangling Reasoning and Prediction (CVPR 2026 · similarity 0.86)
- UGround: Towards Unified Visual Grounding with Unrolled Transformers (ICML 2026 · similarity 0.86)
- Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMs (NeurIPS 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
